It is generally believed by those undertaking research in the fundamen
tal aspects of geophysical fluid dynamics and meteorology that their r
esults contribute to the improvements to numerical weather prediction
and in practical weather forecasting. However, the techniques whereby
the appropriate research results are selected and incorporated into th
e numerical models are not widely known, particularly the methods for
representing the phenomena whose horizontal scale is less than that of
the grid boxes.(say, 50 km). Some accounts of numerical weather predi
ction imply that the representation of subgrid-scale phenomena is form
ally similar to classical physics. In fact, atmospheric motions on the
se scales are not like molecular motions in an ideal gas, but show con
siderable structure, approximating to combinations of various idealize
d states. Great skill and experience in this specialized activity has
been applied to deciding on these states, finding physical criteria fo
r defining them and then modelling the relevant phenomena occurring on
this scale. In this paper, we focus on a restricted range of phenomen
a associated with stably stratified flows, notably mountain waves, con
vection and clouds, and boundary layer phenomena. This category provid
es many examples of structures which need to be considered in detail t
o reconstruct the large-scale picture accurately, as well as in local
forecasting.